Re: linkonce sections, DWARF2 EH, and the ppc failures

[Alan Modra <amodra at bigpond dot net dot au>]

On Wed, Jun 20, 2001 at 08:51:09AM -0700, Richard Henderson wrote:
> On Wed, Jun 20, 2001 at 03:42:03PM +0930, Alan Modra wrote:
> > Why do you think there would be one header per output file?  The linker
> > is just stitching the .eh_frame* info back together in the order given
> > in input files.  If you have one header per .o, they should all appear
> > in the output.
> 
> No, all the .eh_frame1 sections will wind up together,
> then all the .eh_frame2 sections, etc.

Ah, but I was proposing a change to the linker scripts.

cat > z1.s <<EOF
 .section .data1
 .byte 1

 .section .data2
 .byte 2

 .section .data3
 .byte 3

 .section .gnu.linkonce.d4
 .byte 4
EOF

cat > z2.s <<EOF
 .section .data1
 .byte 0x11

 .section .data2
 .byte 0x12

 .section .data3
 .byte 0x13

 .section .gnu.linkonce.d4
 .byte 0x14
EOF

cat > z.lnk <<EOF
OUTPUT_FORMAT("elf32-i386", "elf32-i386",
              "elf32-i386")
OUTPUT_ARCH(i386)
SECTIONS
{
  .data    :
  {
    *(.data* .gnu.linkonce.d*)
  }
}
EOF

as -o z1.o z1.s
as -o z2.o z2.s
ld -T z.lnk -o z z1.o z2.o
objdump --full z

z:     file format elf32-i386

Contents of section .data:
 0000 01020300 11121300 04                 .........       

which almost works.  There are two problems: padding, and the fact that
the linkonce section gets reordered.

Alan